Log Normal Deconvolution Of Laurdan Fluorescence Spectra A Tool To Asses Lipid Membrane Fluidity [608817]

LOG-NORMAL DECONVOLUTION OF LAURDAN
FLUORESCENCE SPECTRA – A TOOL TO ASSESS LIPID
MEMBRANE FLUIDITY
B. ZORILA 1,2 , Mihaela BACALUM 1, A. I. POPESCU 2, M. RADU 1
1Department of Life and Environmental Physics, “Hori a Hulubei” National Institute of Physics and
Nuclear Engineering, M ăgurele, Romania, E-mail: [anonimizat] , [anonimizat] ,
[anonimizat]
2Department of Electricity, Solid Physics and Biophy sics, Faculty of Physics, University of Bucharest,
Măgurele, Romania, E-mail: prof.aurel.popescu@gmail.c om
Abstract. Deconvolution of complex steady-state flu orescence spectra is a key subject in
analytical fluorescence spectroscopy. The shape of the spectra is generated by the presence in the
analyzed solution of a mixture of several fluoropho res or by a single fluorophore found in different
excited states. The spectra shape of most of the fl uorophores is asymmetric, even in a homogeneous
solution, where only one excited state is presumed to be present. Due to this, fluorescence spectra ca n
be analyzed much better by a log-normal (LN) distri bution than by a Gaussian one. Laurdan is a
membrane fluorescent probe who has the advantage of detecting changes in bilayer phase properties.
Laurdan typical red-shift (~ 50 nm) is observed dur ing the phospholipid phase transition, and is
originating from the probe sensitivity to its envir onment polarity. In this study, we propose a
comparison between Gaussian and LN deconvolution of fluorescence spectra of Laurdan, inserted in
large unilamellar vesicles, prepared from lipids wi th different hydrocarbon tails. We used a new
parameter, namely, the difference of relative areas of the elementary peaks ( ∆Sr) to assess lipid
membrane fluidity. We found that the results give a better characterization of the hydration level of
the environment surrounding Laurdan.
Key words: Laurdan fluorescence, generalized polari zation, Gauss vs. log-normal
deconvolution, lipid membrane fluidity, phospholipi d phase transition.
1. INTRODUCTION
The emission spectra of most fluorophores have a co mplex structure and an
asymmetric shape which can be much better analyzed by a log-normal (LN)
function. But, only a few studies have proposed thi s alternative, and very often, the
fluorescence spectra are analyzed using symmetric f unctions like Gaussian or
Lorentzian functions. The use of the LN function to describe the absorption spectra
was first proposed by Siano and Metzler [1] and lat er on adjusted by Burstein to
analyze the emission spectra of fluorophores [2]. T he procedure was first

2
implemented for the analysis of Tryptophan emission spectrum in proteins [3] and
later modified to allow the analysis of other fluor ophores, like Prodan and
Acrylodan [4]. In our previous study, we addressed this procedure for the first time
to Laurdan (6-dodecanoyl-2-dimethylamino naphthalen e) [5].
The solvatochromic dye, Laurdan, has been intensive ly used in the
exploration of both model and natural membrane prop erties [4, 6-8]. Laurdan 12
carbon aliphatic tail facilitates its insertion int o lipid membranes. The naphthalene
moiety of Laurdan is generally considered to be loc ated at the level of the glycerol
backbone [7, 9] and its dipole moment oriented alon g the normal to the membrane
[6, 7, 10]. Laurdan fluorescence properties depend strongly on the relaxation of the
polar solvent molecules around the increased dipole moment of the probe in the
excited state [11]. When inserted into the lipid me mbranes, the relaxation is due to
the water molecules that can penetrate the lipid me mbrane which depends on the
state of the lipid phase [10, 12]. Thus, the lipid phase of the membrane can be
assessed by Laurdan [13].
Considering this, for a membrane with a higher orde r of lipid packing (in the
gel phase), only a few water molecules can reach th e naphthalene moiety of
Laurdan, resulting in only a slight effect of dipol ar relaxation. Conversely, the fluid
phase is characterized by a poor lipid packing whic h allows more water molecules
to penetrate the membrane at the level of the glyce rol backbone [10, 12]. These can
be observed as a red shift in Laurdan emission maxi mum, which goes from around
440 nm, in the gel phase, to around 490 nm, in the fluid phase [7, 10, 14].
Using the fluorescence intensities of Laurdan at 44 0 nm ( Iblu e) and 490 nm
(Igreen ), the so-called generalized polarization ( GP ) was defined as [13]:

GP = (Iblue – Igreen )/( Iblue + Igreen ) (1)

Theoretically, GP values can range from +1 (no solvent effects) to – 1 (strong
solvent effects) in homogenous solutions (using sol vents covering a large range of
polarity from highly apolar cyclohexane to largely polar alcohols) (for details see
[8]). In the lipid bilayer, it was proved experimen tally that GP goes from 0.6 to –
0.4, irrespective of spectroscopy method [8, 12, 13 , 15] or microscopy
measurements [14, 16-19]. GP values cannot reach the extreme values in the lipi d
membranes because of the spectral overlapping of th e emission bands from the
relaxed and non-relaxed states. Nevertheless, if in stead of using the fluorescence
intensities at 440 and 490 nm, we will use the numb er of emitting molecules from
each state, we could generate a more reliable param eter. Such information can be
provided by decomposition of Laurdan spectra in a s uperposition of two
elementary peaks. Thereafter, we can use the relati ve area of the elementary peaks
(as an indirect measure of the number of molecules emitting from each excited
states) to characterize changes in Laurdan spectra. This is usually done using a sum
of two Gaussian functions to solve the fluorescence emission spectra [20-22].
However, in our previous study, we proposed instead , the use of LN functions to
deconvolute the spectra of Laurdan inserted into mo del lipid membranes (DMPC

3
LUVs) [5]. This method proved to describe better th e complex spectra of Laurdan
in DMPC bilayers and to provide a more feasible pic ture concerning the hydration
of these structures at the level of lipid backbone.
In this article, we extended our previously propose d model [5] to analyze the
spectra of Laurdan incorporated into three other ty pes of liposomes and we
compared the results obtained using the LN decompos ition with those generated
using the Gaussian decomposition.
2. MATERIALS AND METHODES
2.1 Materials
Laurdan was purchased from Invitrogen/Molecular Pro bes (Eugene, OR,
USA). The lipids used, 1,2-ditridecanoyl-sn-glycero -3-phosphocholine (C13) 1,2-
dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) and 1,2-distearoyl-sn-glycero-3-
phosphocholine (DSPC), were purchased from Avanti P olar Lipids (Alabaster, AL,
USA). Na 2HPO 4·2H 2O, KH 2PO 4 anhydrous, and NaCl were purchased from
Sigma-Aldrich and used to prepare the phosphate-buf fered saline (PBS, 10 mM,
pH 7.4).
2.2 LUVs preparation
LUVs with a final lipid concentration of 50 µM were prepared using the
extrusion method according to the Avanti published protocol. Briefly, an
appropriate amount of lipids was dried under nitrog en flow to remove the solvent
and to obtain a lipid film. The lipid film was hydr ated with PBS, heated above the
transition temperature ( Tm) of the lipids and vigorously vortexed to form a
suspension of multilamellar vesicles (MLVs). The ML V suspension was repeatedly
freeze–thawed (5 cycles) and extruded (25 times) th rough a 200-µm filter using a
standard extruder (Avanti Polar Lipids). The extrus ion was performed also at a
temperature above Tm of the lipids, resulting in a suspension of LUVs. Laurdan was
added into the LUV suspension to a final lipid/prob e ratio of 500:1.
2.3 Fluorescence spectroscopy measurements
Steady-state fluorescence measurements were perform ed using a FluoroMax
3 spectrofluorimeter (Horiba Jobin Yvon, Edison, NJ , USA) equipped with a
Peltier thermostated cell holder. The emission spec tra of the Laurdan were
recorded in the range, 1 – 70 °C, depending on the lipids used. The spectra were
recorded in the range, 400 – 600 nm, with the excit ation set at 378 nm. The
recorded spectra were corrected for the spectral se nsitivity of the emission channel

4
of the spectrofluorimeter and for Raman and scatter ing artifacts. All of the
emission recordings have been done at a suspension absorption smaller than 0.05.
Consequently, no correction for the inner effect wa s needed [23]. Before the fitting,
all the spectra were converted in the wavenumber sc ale using the relation I = Iλ×λ2,
where I is the intensity in wavenumber scale, Iλ is the intensity in wavelength scale,
and λ is the wavelength [24].
2.4 Spectra decomposition procedure
Spectra recorded for Laurdan inserted into the lipi d bilayer of LUVs were
fitted with a superposition of two LN functions as reported earlier [5] in a script
written using MatLab 9b software (MathWorks, Natick , MA, USA). For
comparison, the complex spectra in LUVs have been d ecomposed by two Gaussian
functions using the Origin 8.0 software package. GP values were calculated using
Origin 8.0.
3. RESULTS AND DISCUSSIONS
We recorded the emission spectra of Laurdan incorpo rated into pure C13,
DPPC or DSPC LUVs. The recordings were performed a t different temperatures
in such a way that, for each type of lipids used, t heir transition from the gel phase
to the fluid phase was well covered. Thus, for C13 the spectra were recorded
between 1 and 60 oC, for DPPC between 20 and 60 o C, while for DSPC were
between 35 and 70 o C. In addition, the spectra recorded for each type of lipids were
normalized with the values recorded at the first te mperature (Figure 1).

Fig. 1 – Normalized fluorescence emission spectra o f Laurdan inserted into pure C13, DPPC, and
DSPC LUVs, at different temperatures

5
Analyzing the family curves presented in Figure 1 w e can observe that for all
three lipids we have the usual behaviour of the flu orescence emission of Laurdan
incorporated into lipid membranes. Namely, the emis sion maximum shifts from
440 nm, when the lipids are in the gel phase to 490 nm, when the membrane is in
the fluid phase.
To describe the contribution of the non-relaxed and relaxed excited states to
the Laurdan emission we deconvoluted the spectra in a sum of two LN functions
using the procedure described previously [5]. Herei nafter, we present our results
obtained by LN decomposition for all experimental c onditions and we will also
make a comparison with the results obtained by much more utilized Gaussian
decomposition.
In Figure 2, the comparison between the decompositi on of Laurdan spectrum
done by LN and Gaussian functions for three represe ntative spectra recorded on
DPPC LUVs, one at 20 °C (A), the second at 41 °C (B ), and the last one at 60 °C
(C) is presented.

Fig. 2 – Comparison between fitting the spectra of Laurdan inserted into DPPC LUV, at different
temperatures (A : 20 oC, B : 41 oC, and C: 60 oC), using the LN or Gaussian functions. The data are
presented in wavenumber scale
The position of the two peaks originated after the Gaussian decomposition
starts from around 435 nm (the blue peak) and 465 n m (the green peak) and varies
greatly with the increase of temperature, reaching 425 nm and 490 nm respectively
(data not shown) in agreement with previous results [18, 21, 22]. Even though
these studies give an explanation for the increase of the position of the second
peak, due to the presence of more water molecules a round Laurdan, none of them
explains the decrease observed for the first band. This decrease is not supported by
physical evidence, meaning that with the increasing temperature the region around
some of Laurdan molecules becomes more rigid with l ess water molecules
surrounding it. We think that our approach, using t he LN decomposition, can give a
more accurate insight into the hydration level in t he membrane as sensed by
Laurdan molecules.

6
In contrast, the outcome from the LN decomposition are two peaks located
around 440 nm for the blue peak and 490 nm for the green peak, which remain
almost independent of the temperature or lipid comp osition (data not shown). At 20
°C (Figure 2A) the lipids are found in the gel phas e (for DPPC, Tm is 41 °C) and it
is expected that Laurdan molecules emit mostly from a nonrelaxed state. From the
LN analysis, the area of the blue peak, located at approximately 440 nm, is as
expected, higher compared with the one of the green peak. On the other hand, the
peaks obtained after Gaussian decomposition does no t show the same trend. But
even more surprisingly, is that the area of the gre en peak is much larger than the
one of the blue peak. These results mean that even when the bilayer is in the gel
phase, most of the Laurdan molecules emit from a re laxed state. For the spectra
recorded at 41 °C (Figure 2B), where it is expected that the gel and fluid phase of
the lipids co-exist [25], we obtain, from the Gauss ian decomposition,
preponderantly a green peak, whereas from the LN de composition the area of the
two peaks are almost equal. Similar tendencies (the green peak obtained using
Gaussian decomposition is larger than the one obtai ned using LN decomposition)
are also found for all the other analyzed spectra. For the last condition, when the
spectra were recorded at 60 °C (at this temperature , the lipids are in the fluid phase)
we notice that for the Gaussian deconvolution there is almost no contribution
coming from the nonrelaxed state, whereas for the L N decompositions, the
contribution diminishes but it still has an importa nt level.
Analyzing the evolution of the elementary peaks are as for the LN
decomposition, we observed for all tested condition s, that at lower temperatures the
blue peak prevails against the green one. Also, wit h increasing temperatures the
area of the blue peak is reduced while that of the green peak becomes larger, until
they become equal around the transition temperature while, at higher temperatures,
the green peak is the most predominant. These resul ts prove that even at low
temperature, when the lipid membrane is in the gel phase, there are Laurdan
molecules that emit from a relaxed state. This desc ription is very similar with that
already presented for DMPC LUVs [5], proving that t he LN decomposition can be
easily extended to analysis of Laurdan spectrum in various lipid environments.
In more details, the temperature dependence of elem entary peak areas of the
data recorded for the three types of LUVs is presen ted in Figure 3, for both LN and
Gaussian deconvolution. The relative area of the pe ak in the blue channel
(emission from non-relaxed state) ( SrB) is ~ 0.92 for C13 LUVs, ~ 0.96 for DPPC
LUVs, and ~ 0.97 for DSPC LUVs at low temperatures, which decreases with
increasing temperature and reaches ~ 0.25 for all t ypes of vesicles in the fluid
phase, whereas, the SrB values after Gaussian deconvolution starts always from
around 0.35, at low temperatures, and decreases to nearly 0, when the membrane is
in the fluid phase, independent of the type of lipi ds. The relative area of the peak in
the green channel ( SrG) has an opposite variation to the SrB one. In Figure 3, we
observe that the peak areas obtained by LN decompos ition have a point where they
meet. Around the temperature, characteristic to thi s point, the gel and fluid phases
are coexisting in the lipid bilayer. In the case of Gaussian decomposition, the green

7
peak area is higher than that of the blue peak for all the temperatures suggesting
that the fluid phase is more present in the bilayer even at temperatures below main
transition temperature. This description is difficu lt to fit into the accepted models
of the hydration during main transition of the lipi d bilayer as we already proved for
DMPC LUVs [5].

Fig. 3 – Comparison of peak relative area resulted from fitting procedures with LN and Gaussian
functions (A: C13, B: DPPC, and C: DSPC). The error bars are the standard deviations resulting
from at least three repeated measurements for each condition
Analyzing in more details the LN peaks area curves presented in figure 3,
some differences (additionally to the main transiti on temperature) can be noticed
among the curves characterizing the lipids used in this study: the relative area peak
at low and high temperature and the temperature ran ge of the transition. The values
of all these parameters are presented in the Table 1. Additionally to the lipids used
in this report, in Table 1 we added also the DMPC, based on the data reported in
our previous work [5].
Firstly, there are slightly differences among the v alues of SrB obtained for the
LUVs of different lipids at lowest temperature. Thi s parameter monotonically
increases with the length of the fatty acid chains. This effect is not observed at the
highest temperature. Secondly, the transition tempe rature range (obtained fitting
the curve with a sigmoid) is monotonically decreasi ng with the length of the fatty
acid chains.

Table 1
Comparison between SrB at extreme values of temperature and ΔT range obtained from sigmoidal fit
of SrB curves (data from Figure 3 and [5] for DMPC)
Lipid Sr B at T min Sr B at T max ΔT
C13 0.92 0.21 7.45
DMPC 0.95 0.26 5.31
DPPC 0.96 0.26 2.02
DSPC 0.97 0.25 1.57

8

These effects may have a simple explanation when on e compares the length
of Laurdan hydrophobic tail (12 carbons) with those of the lipids where it is
inserted, The DSPC lipid chains have 18 carbons com pared with 16 carbons of
DPPC, 14 of DMPC and 13 of C13. As a result, the lo nger lipids can accommodate
better the Laurdan molecules, that is, less water m olecules reaches them in the case
of membranes composed of lipids with longer aliphat ic tails in the well packaged
gel phase. On the contrary, in the fluid phase, the results indicate that the length of
the lipid chains is not important, so that when the membrane becomes more fluid,
Laurdan “senses” similar environments. These differ ences generated by the length
of the lipid tails can also be observed around the transition temperature. Thus, for
the C13 we notice a wider transition region, while with the increase of the lipid
tails, the transition becomes steeper (Figure 3). S uch an effect cannot be related
only to the lipids themselves, because one of the m ost used methods for analyzing
the main transition (differential scanning calorime try) has not evidenced similar
results [26].
The curves of peak area against the temperature can be fitted using a
sigmoidal function and used to obtain information a bout Tm. The point where the
relative areas become equal represents the experime ntal condition where the gel
and fluid phase of the lipids co-exist and the same number of Laurdan molecules
(in the hypothesis of the same quantum yield for em ission from the relaxed and
non-relaxed state) are found in both of them. This should happen at Tm of the lipids
and using the plots from Figure 2, we can determine Tm for each lipid (Table 2).

Table 2
Tm of lipids obtained using the GP or the parameters resulted by the fitting with LN function,
compared with the ones reported in the literature
Tm / °C C13 DPPC DSPC
GP a 22.01 40.93 53.35
∆Sra 19.27 40.32 53.04
SrB, SrGb 30.83 42.81 54.06
SrG/SrBa 23.00 41.50 53.70
Ref [27] 13.50 41.30 54.50
a – sigmoidal fit
b – intersection of SrB and SrG curves

Using the relative areas of the elementary peaks we can define a parameter
similar to the GP , namely, the difference of relative areas: ∆Sr = SrB – SrG. This
parameter describes Laurdan accessibility to water molecules when inserted into
lipid membranes and depends on the fractions of emi tting molecules in each state,
theoretically ranging from +1 (emission from non-re laxed state only) to -1
(emission from relaxed state only). GP values were also calculated using equation

9
(1). Thus, both ∆Sr and GP values obtained for the recordings of Laurdan inse rted
in the liposomes prepared from C13, DPPC and DSPC a re plotted in Figure 4. The
∆Sr values start from high values, 0.85 for C13, 0.91 for DPPC, and ~ 0.95 for
DSPC and drop to ~ -0.5. Compared with ∆Sr values, GP values are 0.44, 0.45 and
0.48, respectively and decrease to ~ -0.4. As we ca n observe, for all tested
conditions, the range covered by ∆Sr is significantly larger than the one covered by
GP . The larger range suggests that ∆Sr can be a more sensitive parameter in
detecting and characterizing the hydration level in the lipid membranes.

Fig. 4 – Comparison between ∆Sr and GP (A: C13, B: DMPC, C: DPPC, and D: DSPC). The
results are means ± standard deviations on three re petitions of the recordings
The GP values calculated for the LUVs used in this study (Figure 4) are in
agreement with the literature [6]. Previous studies on liposomes showed that Tm of
the lipids can be accurately determined by the temp erature dependence of GP [6,
22]. Thus, both GP and ∆Sr were fitted with a sigmoidal function and we
determined Tm of the lipids used to prepare the LUVs (Table 2). The Tm values
reported in Table 2 for ∆Sr are consistent with those derived from GP curves and
also with those reported in literature [27].
We also calculated the ratio of elementary peak are as ( SrG/SrB), which is
another parameter proposed in the literature to cha racterize the changes in Laurdan
spectrum [22]. These parameters calculated for the LN and Gaussian fitting peaks,
for all the LUVs used, are depicted in Figure 5. Th is parameter has a sharp increase
starting close to the phase transition temperature.

10

Fig. 5 – Relative area ratio dependence on temperatu re for pure C13, DPPC, and DSPC LUVs as
obtained for LN or Gaussian decomposition. The resu lts are means ± standard deviations on three
repetitions of the recordings
3. CONCLUSIONS
In this article, we used a method for analysis Laur dan spectrum presented in a
previous work [5] based on a decomposition of spect ra using two asymmetric LN
functions to evaluate the spectra of Laurdan insert ed into the membrane of pure
LUVs prepared from C13, DPPC or DSPC.
The temperature dependence of LN elementary peak a rea is providing a
more realistic image for the main transition phase comparing with the Gaussian
deconvolution for all the lipids used in this work. All the results are consistent with
our findings on DMPC LUVs [5] proving that LN decom position method is
consistent in describing the Laurdan emission when inserted into lipid bilayers,
The new parameter proposed to characterize the stat e of the liposome
membrane, ∆Sr, is a more sensitive with respect to the hydratio n level sensed by
Laurdan. The peak area values in gel phase and the range of temperature
characteristic for the main transition are dependin g on the length of lipid
hydrophobic chains.
The method used in this paper will be directed to t he analysis of more
complex systems and the changes induced in membrane fluidity by different
factors and also to the interactions between protei ns and lipid bilayers.

Acknowledgments. B. Z. is preparing a PhD at Faculty of Physics, Uni versity of Bucharest.
This work was supported by the Romanian Ministry of Education and Research through Grants:

11
PNII-123/2012, PNII-98/2012, PN-II-ID-PCCE-2011-2-002 7 and PN 09370301.
REFERENCES
1. B. D. Siano and D. E. Metzler, J. Chem. Phys. , 51, 1856-1861 (1969).
2. E. A. Burstein and V. I. Emelyanenko, Photochem. Photobiol Sci. , 64 , 316-320 (1996).
3. E. A. Burstein, S. M. Abornev, and Y. K. Reshetnya k, Biophys. J. , 81 , 1699-1709 (2001).
4. V. I. Emel’ianenko, I. Reshetniak, O.A. Andreev, and E.A. Burstein, Biofizika , 45 , 207-219
(2000).
5. M. Bacalum, B. Zorila, and M. Radu, Anal. Biochem. , 440 , 123-129 (2013).
6. L. A. Bagatolli, T. Parasassi, G. D. Fidelio, and E. Gratton, Photochem. Photobiol. , 70 , 557-
564 (1999).
7. L. A. Bagatolli, Biochim. Biophys. Acta , 1758 , 1541-56 (2006).
8. T. Parasassi, G. De Stasio, G. Ravagnan, R. M. Rus ch, and E. Gratton, Biophys. J. , 60 , 179-
89 (1991).
9. P. Jurkiewicz, A. Olzynska, M. Langner, and M. H of, Langmuir , 22 , 8741-8749 (2006).
10. S. A. Sanchez, M. A. Tricerri, G. Gunther, and E. Gratton, Modern research and
educational topics in microscopy: applications in p hysical/chemical sciences , Formatex
Research Center, 1007-1014 (2007).
11. P. Jurkiewicz, L. Cwiklik, P. Jungwirth, and M. Hof, Biochimie , 94 , 26-32 (2012).
12. T. Parasassi, E.K. Krasnowska, L. Bagatolli, and E. Gratton, J. Fluoresc. , 8, 365-373
(1998).
13. T. Parasassi, G. De Stasio, A. Dubaldo, and E. Gratton, Biophys. J. , 57 , 1179-1186 (1990).
14. L. A. Bagatolli and E. Gratton, Biophys. J. , 78 , 290-305 (2000).
15. T. Parasassi, M. Di Stefano, M. Loiero, G. Rav agnan, and E. Gratton, Biophys. J. , 66 , 120-
32 (1994).
16. J. Kubiak, J. Brewer, S. Hansen, and L. A. Bagato lli, Biophys. J. , 100 , 978-86 (2011).
17. D. M. Owen, C. Rentero, A. Magenau, A. Abu-Siniye h, and K. Gaus, Nat. Prot. , 7, 24-35
(2012).
18. W. M. Yu, P. T. C. So, T. French, and E. Gratton , Biophys. J. , 70 , 626-636 (1996).
19. S. A. Sanchez, G. Gunther, M. A. Tricerri, and E. Gratton, J. Membr. Biol. , 241 , 1-10
(2011).
20. C. C. De Vequi-Suplicy, C. R. Benatti, and M. T. Lam y, J. Fluoresc. , 16 , 431-439 (2006).
21. A. D. Lucio, C. C. Vequi-Suplicy, R. M. Fernandez , and M. T. Lamy, J. Fluoresc. , 20 , 473-
82 (2010).
22. D. Ionescu and C. Ganea, Eur. Biophys. J. , 41 , 307-318 (2012).
23. J. R. Lakowicz, Principles of Fluorescence Spectroscopy , 3rd ed. Berlin, Germany,
Springer (2006).
24. B. Valeur, Molecular Fluorescence: Principles and Applications , Weinheim, Germany,
Wiley-VCH (2002).
25. F. Tokumasu, A. J. Jin, and J. A. Dvorak, J. Electron Microsc. 51 , 1-9 (2002).
26. W. Xie, G. Ludewig, K. Wang, and H. J. Lehmler, Colloids and Surfaces. B, Biointerfaces ,
76 , 128-36 (2010).
27. R. Koynova and M. Caffrey, Biochim. Biophys. Acta , 1376 , 91-145 (1998).